基于流形学习的高动态范围图像质量评价

High Dynamic Range Image Quality Assessment Based on Manifold Learning

高动态范围图像通过更高的对比度和更广的显示亮度增加用户的身临其境感和用户体验质量,因而对其质量评价有着广泛的应用需求。已有研究表明流形可作为大脑感知的一种表述,故提出了基于流形学习的高动态范围图像质量评价方法。该方法将高动态范围图像进行预处理后进行亮度分解和纹理提取,分别对亮度分解和纹理提取后的边缘与纹理特征图像进行流形特征相似度度量;然后考虑色度域扩展,对色度分量直接进行结构相似度度量;将六个特征采用随机森林方法得到图像的最终客观质量值。该方法在高动态范围图像数据库中进行验证,其PLCC与SROCC指标分别达到了0.9238与0.9117,研究结果表明该方法能很好地符合人眼视觉感知特性。

High dynamic range image has a wide cause it can increase the users＇ immersive sense and range of application requirements for its quality evaluation be- the quality of experience by higher contrast and wider display brightness. Previous studies had indicated that manifold could be used as a representation of the brain perception, so a new measurement method based on manifold learning （ML） for high dynamic range image quality assessment was pro- posed. After pre-processing, the edge and texture features of the luminance image were formed by brightness decom- position and texture extraction respectively, which were all measured in manifold feature similarity. And then consider- ing the chrominance domain expansion, the chrominance components were respectively measured in structure similari- ty. At the end, the objective quality evaluation value was obtained from the six features based on the random forest al- gorithm supported. The method is verified in the high dynamic range images database and the test result which PLCC and SROCC indexes reach 0. 9238 and 0. 9117 respectively shows that this method can accord with the human visual perception.